딥러닝 기법을 이용한 연안 양식 시설 탐지의 정확도 평가

Due to declining fish catches caused by rapid climate change and advancements in aquaculture technology, the global demand for aquaculture products is continuously increasing. However, since the reckless expansion of facilities adversely affects the coastal ecosystem and fish stock prices, managing aquaculture facilities through periodic coastal environment monitoring is essential. This study analyzed the detection accuracy of shellfish aquaculture facilities in Gyeongsangnam-do using Sentinel-2 optical imageries and deep learning-based detection methodology. The DeepLabv3+, ResUNet++, and Attention U-Net networks were applied, and as a result, Attention U-Net showed the best detection performance with F1 score of 0.8708 and Intersection over Union 0.7708. The detection methodology presented in this study allows periodic observation of aquaculture facilities affected by sea currents and suspending matters. Also, it may apply to detecting various aquaculture species, showing high potential for expansion to wider areas. Therefore, the aquaculture facility information derived through this study is expected to be useful for future policy decisions regarding marine spatial utilization.